Research Scientist, Manipulation Evals
Non-profit AI safety research and community-building lab that researches frontier-AI safety and runs remote research sprints and fellowships.
Funding history
About Apart Research
Apart Research conducts AI safety research, including work on frontier-model manipulation and safety evaluations, and develops researchers through open research sprints and fellowship programs.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will own the end-to-end development of harmful-manipulation evaluations for frontier AI models. You will prioritize risk scenarios, design rigorous methods, run evaluations using coding agents, analyze results, and communicate findings clearly for regulatory and academic audiences. You will conduct literature reviews, investigate model behavior, red-team evaluations, and produce public deliverables such as papers, benchmarks, and technical reports.
Requirements
- Hands-on evaluation experience and strong references
- Experience using coding agents such as Claude Code or Codex
- At least one co-authored paper at a top AI/ML venue, or impactful workshop papers or preprints
- Track record of safety-relevant empirical work in evaluations, benchmarks, adversarial robustness, or alignment
- Strong Python and software engineering skills
- Experience with evaluation frameworks such as Inspect and LLM APIs
- Ability to analyze large, noisy datasets
- Technical writing ability for regulator and academic audiences
- Ability to independently own open-ended research mandates
Responsibilities
- Develop and prioritize harmful-manipulation risk scenarios
- Build and run end-to-end evaluations against frontier models using coding agents
- Design rigorous evaluation methods for AI manipulation
- Conduct literature reviews on manipulation, deception, and persuasion evaluations
- Investigate model behavior and identify manipulation patterns in transcripts
- Convert qualitative observations into quantitative measures
- Red-team your own and others' evaluations
- Deliver new research results at least weekly
- Write technical findings for regulatory and academic audiences
- Co-author papers, benchmark releases, and technical reports
- Engage directly with the EU AI Office
Benefits
- Fully remote work in EU-compatible time zones
- Travel budget for conferences, EU institutional meetings, and coworking across Europe
- Learning and development budget
- Equipment budget
- Home-office budget
Hiring Process
Application form → unpaid work sample → reference checks → in-depth interviews with Research Scientists and CEO → possible paid remote work trial → offer.